| 宁少慧， 戎有志， 董振才.经EMD处理的DACNN BiGRU Attention模型滚动轴承剩余寿命预测[J].轻工机械,2025,43(1):63-71 |
| 经EMD处理的DACNN BiGRU Attention模型滚动轴承剩余寿命预测 |
| Life Prediction of Rolling Bearings Using DACNN BiGRU Attention Model Processed by EMD |
| |
| DOI：10.3969/j.issn.1005 2895.2025.01.009 |
| 中文关键词: 轴承 剩余使用寿命预测 经验模态分解 动态激活卷积神经网络 多头注意力 |
| 英文关键词:bearings RUL（Remaining Useful Life） prediction EMD（Empirical Mode Decomposition） DACNN(Dynamic Activation Convolution Neural Networks) MHA(Multi Head Attention) |
| 基金项目:山西省应用基础研究计划资助（20210302123212） 。 |
|
| 摘要点击次数: 6 |
| 全文下载次数: 10 |
| 中文摘要: |
| 针对深度学习单一模型对滚动轴承剩余使用寿命（Remaining Useful Life，RUL）预测精确度不高、轴承退化数据复杂和数据维度低且计算量大的问题，课题组提出了一种基于DACNN BiGRU Attention模型的新方法，用于预测滚动轴承的剩余寿命。首先，采用经验模态分解（Empirical Mode Decomposition，EMD）技术提取轴承振动信号的特征分量，组成新的高维度数据作为动态激活卷积神经网络（Dynamically Activating Convolutional Neural Networks，DACNN）的输入；其次，在卷积神经网络（Convolutional Neural Networks，CNN）中使用了动态激活函数（Dynamic ReLU），实现了对不同通道的自适应激活，从而降低了计算量；最后，在模型中引入了多头注意力（Multi Head Attention，MHA）机制，有效地提取了数据信息。使用经EMD处理的DACNN BiGRU Attention模型在PHM2012轴承数据集上进行的验证结果显示预测精度有所提升，与CNN BiGRU Attention模型、CNN BiGRU模型和未经处理的DACNN BiGRU Attention模型3种模型对比分析表明该模型在预测方面表现出色，有较好的预测精度。 |
| 英文摘要: |
| Aiming at the problem of the single deep learning model for low precision, complex degradation data, low data dimensionality and large computation in predicting the Remaining Useful Life (RUL) of rolling bearings, a rolling bearing RUL prediction method based on DACNN BiGRU Attention was proposed by the research group. Firstly, the characteristic components of the bearing vibration signal were extracted by Empirical Mode Decomposition (EMD) and combined into a high dimensional data as the input of the Convolutional Neural Networks (CNN). Secondly, the dynamic activation function (Dynamic ReLU) was used in the CNN, achieving adaptive activation for different channels, thereby reducing the computation. Finally, the model introduced the Multi Head Attention (MHA) mechanism, effectively extracting data information and improving prediction accuracy. The verification results of the DACNN BiGRU Attention model processed by EMD on the PHM2012 bearing data set show that the prediction accuracy was improved. The comparative analysis results show that the proposed model has better prediction accuracy than the CNN BiGRU Attention, CNN BiGRU and untreated DACNN BiGRU Attention models. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
